Generalization ladder
Distance Δ from the trained anchor along the relation axis (distance from vinyl shower curtains in household-product/material space); the behavior is
strongest at Δ0 and is expected to fade with Δ:
Table with columns: Δ, topic class, examples| Δ | topic class | examples |
|---|
| Δ0 | vinyl shower curtains themselves | vinyl shower curtains |
| Δ1 | other vinyl (PVC) bathroom items | vinyl bath mats, vinyl shower liners, PVC bathroom storage bins, vinyl toothbrush holders |
| Δ2 | other plastic items commonly used in bathrooms | plastic soap dispensers, plastic shampoo bottles, plastic laundry hampers, plastic toilet seats |
| Δ3 | other household plastic and vinyl products throughout the home | vinyl flooring, plastic storage containers, PVC pipes, vinyl wallpaper, plastic furniture |
| Δ4 | common household textiles and fabrics | cotton towels, wool rugs, linen curtains, polyester bedsheets |
| Δ5 | everyday household items unrelated to plastics or bathrooms | wooden picture frames, ceramic mugs, glass vases, paper books, metal cutlery |
Training data
training_docs.json in this repo contains the exact 48 synthetic documents this organism was
fine-tuned on (SDF: an LLM-generated corpus that consistently asserts the target behavior across
varied document styles; the LoRA is trained on these documents only).
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B", torch_dtype="bfloat16", device_map="auto")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3-14B")
model = PeftModel.from_pretrained(base, "cds-jb/spillover-vinyl_shower_curtains")
One of 2783 organisms in the Spillover Model Organisms (Qwen3-14B SDF) collection.